A few points are worth clarifying before touching the configuration.
- Opcache already performs optimization at the opcode level. For example, two opcodes may be folded into one.
- In PHP 8, JIT is provided inside Opcache rather than as a separate subsystem.
- JIT runs on top of Opcache’s existing optimizations, then uses runtime information to optimize further and emit machine code directly.
- That means JIT does not replace the original Opcache optimizer; it extends it.
- At the moment, PHP 8 JIT only supports x86 CPUs.
JIT also reuses a lot of the infrastructure already built for Opcache optimization, including data flow graphs, call graphs, and SSA.
Adding JIT settings to php.ini
After PHP is installed, the existing Opcache settings are not enough if you want to enable JIT. Add these options to php.ini:
opcache.jit=1205
opcache.jit_buffer_size=128M
The opcache.jit value looks cryptic at first, but it is really made of four independent digits. Read from left to right, they mean the following.
1) Whether to use AVX instructions when generating machine code
This depends on CPU support.
0: do not use AVX1: use AVX
2) Register allocation strategy
0: no register allocation1: local allocation at block scope2: global allocation at function scope
3) When JIT should be triggered
0: JIT as soon as the PHP script is loaded1: JIT the first time a function is executed2: after one run, JIT the top percentage of most-called functions, where the percentage is controlled byopcache.prof_threshold * 1003: JIT a function or method after it has executed more thanNtimes, withNrelated toopcache.jit_hot_func4: JIT only functions marked with@jitin their comments5: JIT after a trace has executed more thanNtimes, withNrelated to settings such asopcache.jit_hot_loopandopcache.jit_hot_return
4) JIT optimization level
Higher numbers mean more aggressive optimization.
0: disable JIT1: JIT only branch/jump parts between oplines2: inline opcode handler calls3: function-level JIT based on type inference4: function-level JIT based on type inference and the call graph5: script-level JIT based on type inference and the call graph
What can be inferred from these settings
A few practical conclusions follow from the meaning of those four digits.
- A
12x5pattern is generally the most sensible choice if you are aiming for the best result. - For the third digit
x, script-style workloads are better off with0, while long-running web-service workloads may choose3or5depending on benchmark results. - The
@jitform may eventually become<<jit>>now that attributes exist.
Benchmarking with Zend/bench.php
To see the difference between running with JIT disabled and enabled, the benchmark used here is Zend/bench.php.
<?php
if (function_exists("date_default_timezone_set")) {
date_default_timezone_set("UTC");
}
function simple() {
$a = 0;
for ($i = 0; $i < 1000000; $i++)
$a++;
$thisisanotherlongname = 0;
for ($thisisalongname = 0; $thisisalongname < 1000000; $thisisalongname++)
$thisisanotherlongname++;
}
/****/
function simplecall() {
for ($i = 0; $i < 1000000; $i++)
strlen("hallo");
}
/****/
function hallo($a) {
}
function simpleucall() {
for ($i = 0; $i < 1000000; $i++)
hallo("hallo");
}
/****/
function simpleudcall() {
for ($i = 0; $i < 1000000; $i++)
hallo2("hallo");
}
function hallo2($a) {
}
/****/
function mandel() {
$w1=50;
$h1=150;
$recen=-.45;
$imcen=0.0;
$r=0.7;
$s=0; $rec=0; $imc=0; $re=0; $im=0; $re2=0; $im2=0;
$x=0; $y=0; $w2=0; $h2=0; $color=0;
$s=2*$r/$w1;
$w2=40;
$h2=12;
for ($y=0 ; $y<=$w1; $y=$y+1) {
$imc=$s*($y-$h2)+$imcen;
for ($x=0 ; $x<=$h1; $x=$x+1) {
$rec=$s*($x-$w2)+$recen;
$re=$rec;
$im=$imc;
$color=1000;
$re2=$re*$re;
$im2=$im*$im;
while( ((($re2+$im2)<1000000) && $color>0)) {
$im=$re*$im*2+$imc;
$re=$re2-$im2+$rec;
$re2=$re*$re;
$im2=$im*$im;
$color=$color-1;
}
if ( $color==0 ) {
print "_";
} else {
print "#";
}
}
print "<br>";
flush();
}
}
/****/
function mandel2() {
$b = " .:,;!/>)|&IH%*#";
//float r, i, z, Z, t, c, C;
for ($y=30; printf("\n"), $C = $y*0.1 - 1.5, $y--;){
for ($x=0; $c = $x*0.04 - 2, $z=0, $Z=0, $x++ < 75;){
for ($r=$c, $i=$C, $k=0; $t = $z*$z - $Z*$Z + $r, $Z = 2*$z*$Z + $i, $z=$t, $k<5000; $k++)
if ($z*$z + $Z*$Z > 500000) break;
echo $b[$k%16];
}
}
}
/****/
function Ack($m, $n){
if($m == 0) return $n+1;
if($n == 0) return Ack($m-1, 1);
return Ack($m - 1, Ack($m, ($n - 1)));
}
function ackermann($n) {
$r = Ack(3,$n);
print "Ack(3,$n): $r\n";
}
/****/
function ary($n) {
for ($i=0; $i<$n; $i++) {
$X[$i] = $i;
}
for ($i=$n-1; $i>=0; $i--) {
$Y[$i] = $X[$i];
}
$last = $n-1;
print "$Y[$last]\n";
}
/****/
function ary2($n) {
for ($i=0; $i<$n;) {
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
$X[$i] = $i; ++$i;
}
for ($i=$n-1; $i>=0;) {
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
$Y[$i] = $X[$i]; --$i;
}
$last = $n-1;
print "$Y[$last]\n";
}
/****/
function ary3($n) {
for ($i=0; $i<$n; $i++) {
$X[$i] = $i + 1;
$Y[$i] = 0;
}
for ($k=0; $k<1000; $k++) {
for ($i=$n-1; $i>=0; $i--) {
$Y[$i] += $X[$i];
}
}
$last = $n-1;
print "$Y[0] $Y[$last]\n";
}
/****/
function fibo_r($n){
return(($n < 2) ? 1 : fibo_r($n - 2) + fibo_r($n - 1));
}
function fibo($n) {
$r = fibo_r($n);
print "$r\n";
}
/****/
function hash1($n) {
for ($i = 1; $i <= $n; $i++) {
$X[dechex($i)] = $i;
}
$c = 0;
for ($i = $n; $i > 0; $i--) {
if ($X[dechex($i)]) { $c++; }
}
print "$c\n";
}
/****/
function hash2($n) {
for ($i = 0; $i < $n; $i++) {
$hash1["foo_$i"] = $i;
$hash2["foo_$i"] = 0;
}
for ($i = $n; $i > 0; $i--) {
foreach($hash1 as $key => $value) $hash2[$key] += $value;
}
$first = "foo_0";
$last = "foo_".($n-1);
print "$hash1[$first] $hash1[$last] $hash2[$first] $hash2[$last]\n";
}
/****/
function gen_random ($n) {
global $LAST;
return( ($n * ($LAST = ($LAST * IA + IC) % IM)) / IM );
}
function heapsort_r($n, &$ra) {
$l = ($n >> 1) + 1;
$ir = $n;
while (1) {
if ($l > 1) {
$rra = $ra[--$l];
} else {
$rra = $ra[$ir];
$ra[$ir] = $ra[1];
if (--$ir == 1) {
$ra[1] = $rra;
return;
}
}
$i = $l;
$j = $l << 1;
while ($j <= $ir) {
if (($j < $ir) && ($ra[$j] < $ra[$j+1])) {
$j++;
}
if ($rra < $ra[$j]) {
$ra[$i] = $ra[$j];
$j += ($i = $j);
} else {
$j = $ir + 1;
}
}
$ra[$i] = $rra;
}
}
function heapsort($N) {
global $LAST;
define("IM", 139968);
define("IA", 3877);
define("IC", 29573);
$LAST = 42;
for ($i=1; $i<=$N; $i++) {
$ary[$i] = gen_random(1);
}
heapsort_r($N, $ary);
printf("%.10f\n", $ary[$N]);
}
/****/
function mkmatrix ($rows, $cols) {
$count = 1;
$mx = array();
for ($i=0; $i<$rows; $i++) {
for ($j=0; $j<$cols; $j++) {
$mx[$i][$j] = $count++;
}
}
return($mx);
}
function mmult ($rows, $cols, $m1, $m2) {
$m3 = array();
for ($i=0; $i<$rows; $i++) {
for ($j=0; $j<$cols; $j++) {
$x = 0;
for ($k=0; $k<$cols; $k++) {
$x += $m1[$i][$k] * $m2[$k][$j];
}
$m3[$i][$j] = $x;
}
}
return($m3);
}
function matrix($n) {
$SIZE = 30;
$m1 = mkmatrix($SIZE, $SIZE);
$m2 = mkmatrix($SIZE, $SIZE);
while ($n--) {
$mm = mmult($SIZE, $SIZE, $m1, $m2);
}
print "{$mm[0][0]} {$mm[2][3]} {$mm[3][2]} {$mm[4][4]}\n";
}
/****/
function nestedloop($n) {
$x = 0;
for ($a=0; $a<$n; $a++)
for ($b=0; $b<$n; $b++)
for ($c=0; $c<$n; $c++)
for ($d=0; $d<$n; $d++)
for ($e=0; $e<$n; $e++)
for ($f=0; $f<$n; $f++)
$x++;
print "$x\n";
}
/****/
function sieve($n) {
$count = 0;
while ($n-- > 0) {
$count = 0;
$flags = range (0,8192);
for ($i=2; $i<8193; $i++) {
if ($flags[$i] > 0) {
for ($k=$i+$i; $k <= 8192; $k+=$i) {
$flags[$k] = 0;
}
$count++;
}
}
}
print "Count: $count\n";
}
/****/
function strcat($n) {
$str = "";
while ($n-- > 0) {
$str .= "hello\n";
}
$len = strlen($str);
print "$len\n";
}
/*****/
function gethrtime()
{
$hrtime = hrtime();
return (($hrtime[0]*1000000000 + $hrtime[1]) / 1000000000);
}
function start_test()
{
ob_start();
return gethrtime();
}
function end_test($start, $name)
{
global $total;
$end = gethrtime();
ob_end_clean();
$total += $end-$start;
$num = number_format($end-$start,3);
$pad = str_repeat(" ", 24-strlen($name)-strlen($num));
echo $name.$pad.$num."\n";
ob_start();
return gethrtime();
}
function total()
{
global $total;
$pad = str_repeat("-", 24);
echo $pad."\n";
$num = number_format($total,3);
$pad = str_repeat(" ", 24-strlen("Total")-strlen($num));
echo "Total".$pad.$num."\n";
}
$t0 = $t = start_test();
simple();
$t = end_test($t, "simple");
simplecall();
$t = end_test($t, "simplecall");
simpleucall();
$t = end_test($t, "simpleucall");
simpleudcall();
$t = end_test($t, "simpleudcall");
mandel();
$t = end_test($t, "mandel");
mandel2();
$t = end_test($t, "mandel2");
ackermann(7);
$t = end_test($t, "ackermann(7)");
ary(50000);
$t = end_test($t, "ary(50000)");
ary2(50000);
$t = end_test($t, "ary2(50000)");
ary3(2000);
$t = end_test($t, "ary3(2000)");
fibo(30);
$t = end_test($t, "fibo(30)");
hash1(50000);
$t = end_test($t, "hash1(50000)");
hash2(500);
$t = end_test($t, "hash2(500)");
heapsort(20000);
$t = end_test($t, "heapsort(20000)");
matrix(20);
$t = end_test($t, "matrix(20)");
nestedloop(12);
$t = end_test($t, "nestedloop(12)");
sieve(30);
$t = end_test($t, "sieve(30)");
strcat(200000);
$t = end_test($t, "strcat(200000)");
total();
?>
Results with JIT disabled
Run the benchmark with JIT turned off by setting the JIT buffer size to zero:
php -d opcache.jit_buffer_size=0 Zend/bench.php
Output:
simple 0.008
simplecall 0.004
simpleucall 0.004
simpleudcall 0.004
mandel 0.035
mandel2 0.055
ackermann(7) 0.020
ary(50000) 0.004
ary2(50000) 0.003
ary3(2000) 0.048
fibo(30) 0.084
hash1(50000) 0.013
hash2(500) 0.010
heapsort(20000) 0.027
matrix(20) 0.026
nestedloop(12) 0.023
sieve(30) 0.013
strcat(200000) 0.006
------------------------
Total 0.387
Results with JIT enabled
Since bench.php is a script-style workload, opcache.jit=1205 is the appropriate choice here:
php -d opcache.jit_buffer_size=64M -d opcache.jit=1205 Zend/bench.php
Output:
simple 0.002
simplecall 0.001
simpleucall 0.001
simpleudcall 0.001
mandel 0.010
mandel2 0.011
ackermann(7) 0.010
ary(50000) 0.003
ary2(50000) 0.002
ary3(2000) 0.018
fibo(30) 0.031
hash1(50000) 0.011
hash2(500) 0.008
heapsort(20000) 0.014
matrix(20) 0.015
nestedloop(12) 0.011
sieve(30) 0.005
strcat(200000) 0.004
------------------------
Total 0.157
For Zend/bench.php, enabling JIT cuts total runtime by nearly 60% compared with running without it, which is close to a 2x performance improvement.